El 09/05/2017 a las 15:46, Ion Gaztañaga via Boost escribió:
On 09/05/2017 13:39, Joaquin M López Muñoz via Boost wrote:
typeid is used extensively throughout the library, both the static (typeid(T)) and the dynamic version (typeid(x)) --the latter only in base_collection and in one spot in boost/poly_collection/detail/segment.hpp. std::type_index is part of the public interface. Boost.TypeIndex could be in principle optionally used for RTTI-less scenarios, but I don't really know whether this is a common enough / requested scenario.
[...]
Basic examples in the documentation are based on games concepts (sprite, warrior, goblin, ...) and usually in game development builds RTTI is disabled although polymorphism is widely used. My comment is about making the runtime-identification configurable by the user
I can imagine an application where all classes are known and a custom typeid can be generated and stored in the abstract base class of every type, so that it can be retrieved via an inline function, without noticeable run-time overhead. It could just cache a std::type_info pointer.
I think a lot of game engines use their own RTTI framework that could be easily plugged into PolyCollection (maybe as an additional TypeIdTraits template parameter). Something like:
class typeid_traits { typedef /**/ type_info_t; typedef /**/ type_index_t;
template<class T> static type_info_t typeid_of(T *p);
//Usage: const type_info_t &ti = typeid_of<MyClass>(); template<class T> static type_info_t typeid_of(); };
type_index_t should be constructible from type_info, hashable and comparable.
This type of abstraction could support Boost.TypeInfo and many other approaches. Anyone sees this customization useful?
I'm no expert in game development so I can't really assess how common this custom typeid thing is. In any case, if the community deems this worth including I can definitely work on it; one thing I don't get about your proposed design is the insistence on the type_info_t/type_index_t distinction: to the best of my knowledge std::type_index should have been what typeid() returned and was added as a patch over less useful typeinfo so as not to break backwards compatibility, so here we can just as well learn from the past and simply use type_index_t (with typeid_of directly returning type_index_'s rather than type_info_t's), right? Two further doubts: * How does this combine with usage of Boost.TypeIndex? I mean, can usage of Boost.TypeIndex be shoehorned into this typeid_traits framework as an alternative RTTI system (among say others custom-designed by game developers)? * function_collection and any_collection share 99% of the code with base_collection (the three containers are basically the same with different so-called "model" policies), so this traits thing could extend almost automatically to them. But I fail to see what the use of it would be. Except when the selected traits are those of Boost.TypeIndex... well, possibilities abound here. Joaquín M López Muñoz